{
"data": {
"frame_length": 0.08,
"input_roles": [
"user",
"User"
],
"output_roles": [
"agent",
"Assistant",
"assistant",
"Agent"
],
"source_sample_rate": 16000,
"target_sample_rate": 22050,
"train_ds": {
"batch_duration": 250,
"bucket_buffer_size": 1000,
"bucket_duration_bins": [
1,
6,
11,
16,
21,
26,
31,
36,
41,
46,
51,
56,
61,
66,
71,
76,
81,
86,
91,
96,
101,
106,
111,
116,
121,
126,
131,
136,
141,
146
],
"max_duration": 142.39029,
"multi_config": true,
"num_workers": 1,
"real_conv_data": {
"batch_duration": 250,
"buffer_size": 1000,
"duration_bins": [
1,
6,
11,
16,
21,
26,
31,
36,
41,
46,
51,
56,
61,
66,
71,
76,
81,
86,
91,
96,
101,
106,
111,
116,
121,
126,
131,
136,
141,
146
],
"input_cfg": "",
"max_duration": 142.39029,
"num_buckets": 31,
"num_workers": 1,
"pin_memory": true,
"quadratic_duration": 20,
"seed": "trng",
"shuffle_buffer_size": 1000,
"use_bucketing": true
},
"sample_rate": 22050,
"sampler_fusion": "randomized_round_robin",
"sampler_weights": {
"real_conv_data": 0.05
},
"seed": 42,
"shard_seed": "trng",
"shuffle": true,
"use_bucketing": true
},
"validation_ds": {
"batch_size": 1,
"datasets": {
"ultrachat": {
"shar_path": ""
}
},
"sample_rate": 22050,
"seed": 42,
"shard_seed": "randomized"
}
},
"exp_manager": {
"create_checkpoint_callback": false,
"create_tensorboard_logger": false,
"create_wandb_logger": false,
"exp_dir": null,
"explicit_log_dir": "",
"name": "",
"resume_from_checkpoint": null,
"resume_if_exists": false,
"resume_ignore_no_checkpoint": true,
"use_datetime_version": true,
"wandb_logger_kwargs": {
"name": "",
"project": "",
"resume": false
}
},
"hf_export_dir": "",
"model": {
"audio_save_path": "",
"inference_speaker_name": "Aria",
"inference_speaker_reference": "",
"lr_scheduler": {
"_target_": "nemo.core.optim.lr_scheduler.InverseSquareRootAnnealing",
"max_steps": 1,
"min_lr": 1e-06,
"warmup_steps": 2500
},
"optimizer": {
"_target_": "torch.optim.AdamW",
"betas": [
0.9,
0.98
],
"foreach": true,
"lr": 0.0003,
"weight_decay": 0
},
"scoring_asr": "stt_en_fastconformer_transducer_large",
"speech_generation": {
"data": {
"audio_prompt_duration": 3.0,
"frame_length": 0.08,
"source_sample_rate": 22050,
"target_sample_rate": 22050
},
"exp_manager": {
"explicit_log_dir": " "
},
"model": {
"audio_save_path": "",
"bos_token": "",
"codec_config": {
"base_hidden_size": 384,
"channel_mult": [
1,
2,
4
],
"codebook_size": 1024,
"groups": 1,
"hop_length": 4,
"kernel_size": 7,
"latent_size": 512,
"n_fft": 16,
"num_blocks": 3,
"num_quantizers": 31,
"rates": [
7,
7,
9
],
"wav_to_token_ratio": 1764
},
"context_hidden_mask_exactly_as_eartts": false,
"eos_token": "",
"freeze_params": [
"^audio_codec\\..+$",
"^embed_tokens\\..+$"
],
"inference_force_speech_silence_on_eos": true,
"inference_guidance_enabled": true,
"inference_guidance_scale": 0.2,
"inference_noise_scale": 0.001,
"inference_top_p_or_k": 0.95,
"lr_scheduler": {
"_target_": "nemo.core.optim.lr_scheduler.InverseSquareRootAnnealing",
"max_steps": 1,
"min_lr": 1e-06,
"warmup_steps": 2500
},
"optimizer": {
"_target_": "torch.optim.AdamW",
"betas": [
0.9,
0.98
],
"foreach": true,
"lr": 4e-05,
"weight_decay": 0
},
"pad_token": "",
"pretrained_ae_dir": null,
"pretrained_lm_name": "nvidia/NVIDIA-Nemotron-Nano-9B-v2",
"pretrained_model": "",
"pretrained_tts_model": null,
"prevent_freeze_params": [],
"subword_mask_exactly_as_eartts": false,
"tts_config": {
"backbone_config": {
"attention_dropout": 0.1,
"head_dim": 72,
"hidden_size": 1152,
"intermediate_size": 4608,
"num_attention_heads": 16,
"num_hidden_layers": 28,
"num_key_value_heads": 16,
"sliding_window": 7500,
"use_cache": false
},
"backbone_config_class": null,
"backbone_model_class": null,
"backbone_type": "gemma3_text",
"cas_config": {
"backbone_config": {
"encoder": {
"attention_dropout": 0.1,
"head_dim": 72,
"hidden_size": 1152,
"intermediate_size": 4608,
"num_attention_heads": 16,
"num_hidden_layers": 1,
"num_key_value_heads": 16,
"use_cache": false
},
"is_encoder_decoder": false
},
"backbone_config_class": null,
"backbone_model_class": null,
"backbone_type": "t5gemma",
"pretrained_tokenizer_name": "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
},
"codebook_size": 1024,
"context_hidden_size": null,
"disable_eos_prediction": true,
"exponent": 3.0,
"label_smoothing": 0.01,
"latent_size": 512,
"max_training_rate": 0.8,
"mog_head_config": {
"eps": 1e-06,
"intermediate_size": 4608,
"low_rank": 64,
"min_log_std": -4.0,
"num_layers": 3,
"num_predictions": 1024
},
"num_delay_speech_tokens": 2,
"num_quantizers": 31,
"p_uncond": 0.1,
"quantizer_dropout": 0.5,
"random_target_masking": false,
"use_audio_prompt_frozen_projection": true,
"use_bos_eos_emb": true,
"use_gated_fusion_for_text_audio": true,
"use_subword_flag_emb": true
},
"use_system_prompt": false
}
},
"stt": {
"data": {
"source_sample_rate": 16000,
"target_sample_rate": 22050
},
"exp_manager": {
"explicit_log_dir": " "
},
"model": {
"allow_user_text_in_agent_turn": false,
"asr_loss_weight": 0.0,
"audio_loss_weight": 0,
"audio_save_path": "",
"augment_fc_system_prompt": true,
"backchannel_debug": false,
"backchannel_debug_max_files": 16,
"backchannel_debug_path": "",
"backchannel_file_path": "",
"backchannel_prob": 0.05,
"backchannel_prob_scale": 0.5,
"backchannel_snr_db": 15,
"base_model_name": "backbone",
"bos_token": "",
"cache_key": "cache_params",
"codec_prob": 0.1,
"delay_source_text_by": 15,
"delay_text_channel_by": 2,
"duplex_asr_text_weight": 1.0,
"duplex_function_channel_weight": 2.0,
"duplex_text_channel_weight": 1.0,
"duplex_user_channel_weight": 1.0,
"embed_tokens_name": "embeddings",
"eos_token": "",
"eval_text_turn_taking": true,
"force_align_device": "cuda",
"force_align_user_text": false,
"force_use_noise_augmentation": true,
"freeze_params": [
"^audio_codec\\..+$"
],
"function_loss_weight": 1.0,
"function_token_loss_weight": {
"call": 64.0,
"eotc": 6.0,
"eotr": 3.0,
"pad": 0.3,
"sotc": 6.0
},
"incremental_loading": true,
"lr_scheduler": {
"_target_": "nemo.core.optim.lr_scheduler.InverseSquareRootAnnealing",
"max_steps": 200000,
"min_lr": 5e-06,
"warmup_steps": 2500
},
"mask_sequence_loss": true,
"max_fc_total_tokens": 3000,
"micir_aug_path": "",
"micir_prob": 0.6,
"nemo_text_norm_cache_dir": "",
"nemo_text_norm_input_case": "cased",
"nemo_text_norm_lang": "en",
"nsynth_noise_aug_path": "",
"nsynth_noise_prob": 0.3,
"old_noise_aug_path": "",
"old_noise_max_snr": 60,
"old_noise_min_snr": -30,
"old_noise_prob": 0.5,
"optimizer": {
"_target_": "torch.optim.AdamW",
"betas": [
0.9,
0.98
],
"foreach": true,
"lr": 5e-05,
"weight_decay": 0
},
"override_tokens": {
"bos_token": "",
"eos_token": "",
"pad_token": ""
},
"pad_token": "",
"perception": {
"encoder": {
"_target_": "nemo.collections.asr.modules.ConformerEncoder",
"att_context_size": [
70,
0
],
"att_context_style": "chunked_limited",
"causal_downsampling": true,
"conv_context_size": "causal",
"conv_kernel_size": 9,
"conv_norm_type": "layer_norm",
"d_model": 1024,
"dropout": 0.1,
"dropout_att": 0.1,
"dropout_emb": 0.0,
"dropout_pre_encoder": 0.1,
"feat_in": 128,
"feat_out": -1,
"ff_expansion_factor": 4,
"n_heads": 8,
"n_layers": 24,
"pos_emb_max_len": 5000,
"reduction": null,
"reduction_factor": 1,
"reduction_position": null,
"self_attention_model": "rel_pos",
"stochastic_depth_drop_prob": 0.0,
"stochastic_depth_mode": "linear",
"stochastic_depth_start_layer": 1,
"subsampling": "dw_striding",
"subsampling_conv_channels": 256,
"subsampling_factor": 8,
"untie_biases": true,
"use_bias": false,
"xscaling": false
},
"modality_adapter": {
"_target_": "nemo.collections.speechlm2.modules.perception.IdentityConnector",
"d_model": 1024
},
"output_dim": 4480,
"preprocessor": {
"_target_": "nemo.collections.asr.modules.AudioToMelSpectrogramPreprocessor",
"dither": 1e-05,
"features": 128,
"frame_splicing": 1,
"log": true,
"n_fft": 512,
"normalize": "NA",
"pad_to": 0,
"pad_value": 0.0,
"sample_rate": 16000,
"window": "hann",
"window_size": 0.025,
"window_stride": 0.01
},
"target": "nemo.collections.speechlm2.modules.perception.AudioPerceptionModule"
},
"predict_user_text": false,
"pretrained_asr": "",
"pretrained_llm": "nvidia/NVIDIA-Nemotron-Nano-9B-v2",
"pretrained_s2s_model": "",
"pretrained_weights": false,
"prevent_freeze_params": [],
"roomir_aug_path": "",
"roomir_prob": 0.8,
"scoring_asr": "nvidia/parakeet-tdt-1.1b",
"text_embedding_weight": 1.0,
"text_loss_weight": 1,
"text_to_text_loss_weight": 0.5,
"token_loss_weight": {
"bos": 12.5,
"eos": 7.5,
"pad": 1.0,
"text": 5.0
},
"use_cache_for_nemotron": false,
"use_codec_aug": true,
"use_function_head": true,
"use_mic_ir_aug": true,
"use_nemo_text_norm": true,
"use_nsynth_noise_aug": false,
"use_numbers_norm": false,
"use_old_noise_aug": true,
"use_room_ir_aug": true,
"use_separate_asr_head": false
}
}
},
"trainer": {
"accelerator": "gpu",
"accumulate_grad_batches": 1,
"devices": -1,
"enable_checkpointing": false,
"gradient_clip_val": 1.0,
"limit_train_batches": 1,
"limit_val_batches": 0.0,
"log_every_n_steps": 20,
"logger": false,
"max_steps": 1,
"num_nodes": 1,
"num_sanity_val_steps": 0,
"precision": 32,
"strategy": {
"_target_": "lightning.pytorch.strategies.DDPStrategy",
"find_unused_parameters": true,
"gradient_as_bucket_view": true
},
"use_distributed_sampler": false,
"val_check_interval": 1
},
"_rnnt_merge_info": {
"s2s_source": "",
"rnnt_source": "",
"rnnt_source_checkpoint": "",
"rnnt_decoder_params": 9,
"rnnt_joint_params": 6,
"decoder_config": {
"_target_": "nemo.collections.asr.modules.RNNTDecoder",
"normalization_mode": null,
"random_state_sampling": false,
"blank_as_pad": true,
"prednet": {
"pred_hidden": 640,
"pred_rnn_layers": 2,
"t_max": null,
"dropout": 0.2
},
"vocab_size": 1024
},
"joint_config": {
"_target_": "nemo.collections.asr.modules.RNNTJoint",
"log_softmax": null,
"preserve_memory": false,
"fuse_loss_wer": true,
"fused_batch_size": 4,
"jointnet": {
"joint_hidden": 640,
"activation": "relu",
"dropout": 0.2,
"encoder_hidden": 1024,
"pred_hidden": 640
},
"num_classes": 1024,
"vocabulary": [
"",
"\u2581t",
"\u2581th",
"\u2581a",
"in",
"\u2581the",
"re",
"\u2581w",
"\u2581o",
"\u2581s",
"er",
"at",
"ou",
"nd",
"it",
"is",
"\u2581h",
"\u2581b",
"on",
"\u2581c",
"ing",
"en",
"\u2581to",
"\u2581m",
"\u2581f",
"\u2581p",
"or",
"an",
"es",
"\u2581of",
"\u2581d",
"ed",
"ll",
"\u2581and",
"\u2581I",
"\u2581in",
"\u2581l",
"ar",
"\u2581y",
"\u2581g",
"as",
"\u2581you",
"om",
"\u2581n",
"ic",
"ve",
"al",
"ion",
"us",
"\u2581be",
"ow",
"le",
"\u2581wh",
"\u2581e",
"ot",
"ut",
"\u2581it",
"\u2581is",
"\u2581we",
"\u2581T",
"\u2581re",
"et",
"\u2581A",
"ent",
"\u2581on",
"\u2581ha",
"ay",
"\u2581S",
"ct",
"\u2581Th",
"ver",
"id",
"ig",
"im",
"ro",
"\u2581for",
"ly",
"\u2581he",
"ke",
"ld",
"se",
"st",
"ch",
"\u2581st",
"all",
"ce",
"ur",
"ith",
"am",
"if",
"ir",
"\u2581go",
"\u2581u",
"\u2581as",
"\u2581was",
"ad",
"\u2581W",
"\u2581k",
"\u2581an",
"ht",
"th",
"\u2581r",
"\u2581are",
"ere",
"\u2581se",
"\u2581do",
"\u2581B",
"\u2581so",
"\u2581sh",
"\u2581not",
"\u2581li",
"od",
"\u2581C",
"ust",
"ill",
"ight",
"ally",
"\u2581And",
"ter",
"\u2581or",
"\u2581me",
"\u2581M",
"ome",
"op",
"\u2581at",
"il",
"\u2581The",
"ould",
"\u2581j",
"ant",
"\u2581So",
"\u2581H",
"ol",
"ain",
"\u2581can",
"\u2581de",
"\u2581ne",
"ore",
"\u2581con",
"\u2581kn",
"ck",
"ul",
"\u2581fr",
"\u2581ab",
"ers",
"ess",
"ge",
"\u2581pro",
"pe",
"ate",
"\u2581su",
"\u2581com",
"\u2581but",
"\u2581all",
"est",
"qu",
"\u2581ex",
"\u2581al",
"ra",
"\u2581O",
"out",
"use",
"very",
"pp",
"\u2581Y",
"\u2581ch",
"ri",
"ist",
"\u2581v",
"\u2581lo",
"ment",
"art",
"\u2581P",
"nt",
"ab",
"\u2581one",
"\u2581N",
"ive",
"\u2581wor",
"ions",
"ort",
"\u2581L",
"\u2581by",
"ich",
"\u2581my",
"ity",
"ok",
"\u2581G",
"res",
"\u2581up",
"un",
"um",
"ea",
"ind",
"and",
"ink",
"el",
"\u2581D",
"em",
"\u2581E",
"os",
"oug",
"\u2581if",
"ca",
"\u2581out",
"\u2581int",
"ie",
"\u2581F",
"\u2581It",
"\u2581his",
"ard",
"\u2581had",
"\u2581tr",
"her",
"our",
"ies",
"ake",
"\u2581R",
"\u2581We",
"\u2581get",
"\u2581don",
"\u2581us",
"ak",
"\u2581pl",
"ect",
"ure",
"ame",
"ast",
"\u2581who",
"ack",
"\u2581le",
"\u2581sa",
"iv",
"ci",
"ide",
"\u2581tim",
"\u2581our",
"ound",
"ous",
"\u2581co",
"\u2581pe",
"ose",
"ud",
"\u2581see",
"ough",
"\u2581man",
"\u2581qu",
"\u2581You",
"so",
"ople",
"\u2581Wh",
"ong",
"ap",
"ther",
"\u2581J",
"are",
"ine",
"\u2581say",
"\u2581im",
"\u2581But",
"ings",
"\u2581has",
"\u2581ag",
"ff",
"\u2581her",
"itt",
"one",
"\u2581en",
"\u2581ar",
"\u2581fe",
"ven",
"\u2581any",
"\u2581mo",
"reat",
"ag",
"\u2581how",
"\u2581cl",
"pt",
"\u2581now",
"own",
"ber",
"\u2581him",
"\u2581act",
"hing",
"ice",
"\u2581no",
"ans",
"iz",
"\u2581fa",
"per",
"pl",
"\u2581te",
"\u2581ad",
"age",
"ree",
"\u2581tw",
"ank",
"\u2581He",
"ple",
"ite",
"ry",
"\u2581U",
"ish",
"ire",
"ue",
"\u2581In",
"\u2581she",
"ble",
"cc",
"nder",
"\u2581way",
"\u2581pr",
"ear",
"\u2581did",
"\u2581po",
"eah",
"\u2581un",
"omet",
"ence",
"ep",
"uch",
"\u2581sp",
"ach",
"og",
"ance",
"able",
"iff",
"sel",
"\u2581got",
"way",
"\u2581gr",
"alk",
"\u2581res",
"ated",
"irst",
"ick",
"ass",
"\u2581two",
"\u2581dis",
"ord",
"\u2581pre",
"ount",
"ase",
"ip",
"ult",
"ical",
"orm",
"ary",
"ace",
"\u2581spe",
"\u2581Ch",
"\u2581thr",
"\u2581imp",
"int",
"\u2581am",
"\u2581off",
"act",
"ia",
"\u2581ro",
"ress",
"\u2581per",
"\u2581fo",
"\u2581br",
"\u2581K",
"vel",
"\u2581gu",
"\u2581bo",
"ang",
"kay",
"ub",
"ign",
"\u2581may",
"ving",
"ces",
"ens",
"cl",
"\u2581lot",
"ru",
"ade",
"\u2581bet",
"\u2581bl",
"\u2581let",
"fore",
"co",
"ild",
"ning",
"xt",
"ile",
"ark",
"self",
"\u2581app",
"ory",
"du",
"\u2581day",
"\u2581St",
"ater",
"\u2581use",
"ys",
"fter",
"\u2581new",
"ious",
"ial",
"he",
"wn",
"ved",
"red",
"\u2581fl",
"iss",
"ody",
"form",
"ian",
"tain",
"\u2581bu",
"\u2581V",
"\u2581rec",
"ty",
"be",
"\u2581sc",
"ors",
"vers",
"\u2581put",
"ife",
"\u2581If",
"we",
"te",
"ject",
"ath",
"ting",
"\u2581rem",
"\u2581acc",
"ull",
"ons",
"\u2581ind",
"\u2581ser",
"\u2581ke",
"ates",
"ves",
"na",
"lic",
"\u2581des",
"\u2581its",
"ful",
"ents",
"erm",
"ac",
"ered",
"ise",
"\u2581sy",
"urn",
"\u2581em",
"oth",
"ual",
"ne",
"ward",
"ib",
"\u2581try",
"\u2581pos",
"nds",
"ft",
"get",
"ph",
"\u2581ob",
"ady",
"igh",
"ood",
"\u2581rel",
"\u2581wr",
"ug",
"ears",
"ail",
"\u2581Now",
"\u2581bit",
"ng",
"\u2581Oh",
"\u2581hel",
"ange",
"\u2581reg",
"\u2581rep",
"\u2581bel",
"\u2581sm",
"ost",
"tern",
"gr",
"\u2581own",
"\u2581end",
"pect",
"ily",
"day",
"ied",
"ific",
"ower",
"\u2581add",
"cess",
"ict",
"ible",
"\u2581bas",
"\u2581i",
"\u2581op",
"cial",
"ular",
"\u2581Be",
"ced",
"\u2581too",
"ks",
"ew",
"mer",
"\u2581ph",
"ob",
"==",
"\u2581la",
"\u2581set",
"\u2581min",
"\u2581sub",
"\u2581gen",
"atch",
"..",
"\u2581inv",
"\u2581As",
"\u2581nat",
"\u2581sl",
"\u2581num",
"av",
"ways",
"\u2581God",
"stem",
"\u2581ac",
"\u2581att",
"\u2581ev",
"\u2581def",
"llow",
"\u2581str",
"lect",
"ars",
"\u2581cr",
"\u2581Is",
"olog",
"les",
"oy",
"\u2581ask",
"\u2581inc",
"body",
"\u2581ent",
"\u2581pol",
"ness",
"ix",
"\u2581why",
"onna",
"\u2581ear",
"\u2581tak",
"\u2581Un",
"ited",
"mun",
"li",
"ute",
"ract",
"\u2581dec",
"uro",
"\u2581mak",
"\u2581fin",
"ween",
"\u2581No",
"arch",
"\u2581bec",
"gan",
"old",
"cy",
"\u2581big",
"\u2581For",
"ren",
"als",
"und",
"\u2581Al",
"\u2581All",
"ss",
"ows",
"\u2581mod",
"ock",
"\u2581id",
"ism",
"cus",
"\u2581gl",
"ably",
"\u2581ass",
"\u2581car",
"ata",
"ppen",
"led",
"\u2581sim",
"\u2581mon",
"ics",
"\u2581giv",
"cept",
"\u2581Mr",
"pan",
"\u2581pub",
"\u2581eff",
"\u2581How",
"ps",
"vern",
"end",
"hip",
"iew",
"ope",
"\u2581An",
"\u2581She",
"\u2581Com",
"ee",
"ures",
"ell",
"ouse",
"cond",
"king",
"oc",
"ues",
"ever",
"\u2581To",
"clud",
"\u2581ins",
"\u2581exp",
"\u2581old",
"\u2581mem",
"\u2581ref",
"\u2581tra",
"\u2581far",
"ave",
"rat",
"\u2581sur",
"ruct",
"rib",
"duct",
"uff",
"\u2581met",
"\u2581sch",
"ince",
"\u2581run",
"ense",
"\u2581cle",
"\u2581==",
"mon",
"ize",
"\u2581ord",
"blem",
"tin",
"\u2581Let",
"ner",
"ond",
"its",
"\u2581cor",
"land",
"\u2581cur",
"\u2581Re",
"\u2581bus",
"\u2581uh",
"air",
"ote",
"ants",
"ason",
"ric",
"\u2581el",
"\u2581cer",
"nce",
"\u2581fam",
"\u2581cap",
"uck",
"ool",
"ried",
"\u2581cou",
"\u2581fun",
"\u2581wom",
"\u2581hum",
"\u2581ty",
"\u2581ap",
"ike",
"\u2581few",
"oney",
"\u2581inf",
"ont",
"ese",
"ook",
"gy",
"uth",
"ulat",
"ieve",
"ized",
"ross",
"\u2581ple",
"\u2581um",
"\u2581val",
"\u2581equ",
"\u2581lea",
"\u2581lar",
"ah",
"eral",
"\u2581ed",
"ared",
"lish",
"arn",
"ds",
"esn",
"\u2581iss",
"\u2581ca",
"ted",
"ices",
"\u2581wee",
"ash",
"\u2581top",
"ten",
"up",
"ts",
"gin",
"con",
"ari",
"\u2581opp",
"osed",
"\u2581eas",
"\u2581ext",
"gg",
"az",
"\u2581Fr",
"ideo",
"izat",
"\u2581men",
"\u2581mom",
"\u2581ret",
"tty",
"rist",
"\u2581gra",
"alth",
"ef",
"\u2581det",
"ax",
"\u2581mat",
"chn",
"ern",
"peri",
"\u2581bre",
"\u2581Sh",
"sw",
"erat",
"\u2581sit",
"ters",
"ale",
"man",
"\u2581sol",
"ork",
"\u2581adv",
"ety",
"\u2581vis",
"\u2581med",
"uc",
"less",
"\u2581unt",
"gram",
"ets",
"ists",
"\u2581ey",
"\u2581col",
"imes",
"\u2581law",
"\u2581pri",
"sid",
"\u2581On",
"\u2581mot",
"ield",
"\u2581Do",
"\u2581At",
"ages",
"amp",
"\u2581art",
"miss",
"\u2581sk",
"alf",
"pr",
"ier",
"\u2581beh",
"\u2581Yes",
"ural",
"ime",
"\u2581wa",
"oks",
"bers",
"ger",
"ient",
"ries",
"...",
"\u2581che",
"\u2581Br",
"ird",
"\u2581Ar",
"\u2581war",
"inat",
"\u2581My",
"ital",
"wh",
"med",
"\u2581pur",
"ully",
"\u2581One",
"\u2581rat",
"ines",
"\u2581Of",
"io",
"\u2581loc",
"ret",
"ctor",
"\u2581leg",
"stit",
"ined",
"ught",
"\u2581dur",
"\u2581es",
"vent",
"aj",
"\u2581bro",
"\u2581saw",
"\u2581sec",
"ream",
"\u2581pop",
"reen",
"\u2581Ind",
"els",
"\u2581yet",
"ired",
"\u2581sw",
"tro",
"oup",
"most",
"pean",
"eds",
"ush",
"oh",
"\u2581Se",
"\u2581tea",
"ann",
"ilit",
"err",
"pend",
"ton",
"ased",
"\u2581aff",
"\u2581mor",
"\u2581dra",
"put",
"\u2581dr",
"ins",
"uat",
"nect",
"cri",
"outh",
"\u2581ra",
"\u2581pay",
"ms",
"\u2581av",
"bs",
"ling",
"\u2581De",
"\u2581Or",
"ove",
"\u2581Can",
"\u2581eng",
"ames",
"ided",
"\u2581Go",
"mitt",
"ode",
"\u2581cre",
"par",
"ides",
"pos",
"\u2581fav",
"\u2581air",
"\u2581New",
"\u2581bad",
"\u2581six",
"vat",
"\u2581pat",
"not",
"\u2581di",
"rop",
"ral",
"orn",
"\u2581par",
"cing",
"\u2581aw",
"orts",
"ox",
"\u2581yes",
"cuss",
"eng",
"ives",
"erms",
"\u2581job",
"mand",
"ying",
"\u2581occ",
"aps",
"ases",
"\u2581Not",
"rent",
"ency",
"att",
"ised",
"vice",
"\u2581Eng",
"\u2581est",
"oked",
"\u2581Q",
"iron",
"idd",
"me",
"unch",
"ane",
"\u2581z",
"br",
"arts",
"\u2581fat",
"ery",
"anks",
"\u2581jo",
"\u2581mar",
"aw",
"ott",
"ards",
"\u2581oh",
"ians",
"\u2581sci",
"row",
"unt",
"ury",
"\u2581abs",
"ergy",
"\u2581Z",
"ump",
"\u2581Am",
"ened",
"angu",
"\u2581Pro",
"icat",
"itch",
"\u2581dri",
"iat",
"\u2581",
"e",
"t",
"o",
"a",
"n",
"i",
"s",
"r",
"h",
"l",
"d",
"u",
"c",
"m",
"y",
"g",
"w",
"f",
"p",
",",
".",
"b",
"v",
"k",
"'",
"I",
"T",
"A",
"S",
"x",
"W",
"j",
"C",
"B",
"M",
"?",
"H",
"O",
"0",
"P",
"q",
"Y",
"N",
"L",
"D",
"1",
"E",
"G",
"z",
"F",
"R",
"-",
"2",
"J",
"U",
"9",
"K",
"5",
"3",
"V",
"=",
"4",
"8",
"6",
"7",
"!",
"%",
":",
"Q",
"Z",
"$",
"X",
"\"",
"&",
"*",
"/",
"\u00a3",
"+",
"\u20ac",
"_",
"^",
"\u00a5"
]
},
"decoder_class": "nemo.collections.asr.modules.rnnt.RNNTDecoder",
"joint_class": "nemo.collections.asr.modules.rnnt.RNNTJoint",
"rnnt_vocab_size": 1024
}
}